Audio & Video Transcription
Transcribe audio and video files from the vault into structured Obsidian notes using read_file to send binary media directly to the model.
Supported Formats
Audio: .wav, .mp3, .aac, .flac, .webm (audio-only)
Video: .mp4, .mpeg, .mov, .flv, .mpg, .webm, .wmv, .3gp
Size limit: 20 MB per file (Gemini inline data limit).
How to Transcribe
- Use
read_file with the path to the audio/video file — the binary data is sent directly to the model for processing.
- Listen to/watch the content and produce a transcription.
- Use
write_file to save the transcription as a markdown note.
Transcription Format
Structure transcriptions as follows:
---
tags:
- transcription
source: "[[original-file.mp3]]"
date: YYYY-MM-DD
duration: "MM:SS" (estimate if possible)
---
# Transcription: [Title]
## Summary
Brief 2-3 sentence summary of the content.
## Transcript
[00:00] Speaker 1: Opening remarks...
[00:45] Speaker 2: Response...
[01:30] Speaker 1: Follow-up...
Guidelines
- Timestamps: Include approximate timestamps in
[MM:SS] format at natural breaks (new speakers, topic changes, pauses).
- Speaker identification: Label distinct speakers as "Speaker 1", "Speaker 2", etc. If names are mentioned, use them after first identification.
- Filler words: Omit excessive filler words (um, uh, like) unless they carry meaning.
- Inaudible sections: Mark unclear audio as
[inaudible] or [unclear].
- Non-speech sounds: Note significant sounds like
[laughter], [applause], [music].
- Summary: Always include a brief summary at the top for quick reference.
- Frontmatter: Link back to the source file using a wikilink.
Tips
- For long recordings, let the user know the transcription may be partial due to the 20 MB size limit. Suggest splitting large files with an external tool.
- If the user asks to "transcribe the recording in this note", use
read_file on the current note first to find embedded audio/video links (e.g., ![[recording.mp3]]), then read_file on the linked file.
- For meeting notes, suggest adding attendees and action items sections after the transcript.
- For podcasts or interviews, suggest adding a "Key Topics" section with timestamps.
1---2name: audio-transcription3description: Transcribe audio and video files into structured notes. Activate this skill when users want to transcribe recordings, meetings, podcasts, voice memos, or any audio/video content in their vault.4metadata:5 source: https://skillmd.com/skills/allenhutchison/audio-transcription6---78# Audio & Video Transcription910Transcribe audio and video files from the vault into structured Obsidian notes using `read_file` to send binary media directly to the model.1112## Supported Formats1314Audio: `.wav`, `.mp3`, `.aac`, `.flac`, `.webm` (audio-only)15Video: `.mp4`, `.mpeg`, `.mov`, `.flv`, `.mpg`, `.webm`, `.wmv`, `.3gp`1617**Size limit:** 20 MB per file (Gemini inline data limit).1819## How to Transcribe20211. Use `read_file` with the path to the audio/video file — the binary data is sent directly to the model for processing.222. Listen to/watch the content and produce a transcription.233. Use `write_file` to save the transcription as a markdown note.2425## Transcription Format2627Structure transcriptions as follows:2829```markdown30---31tags:32 - transcription33source: "[[original-file.mp3]]"34date: YYYY-MM-DD35duration: "MM:SS" (estimate if possible)36---3738# Transcription: [Title]3940## Summary4142Brief 2-3 sentence summary of the content.4344## Transcript4546[00:00] Speaker 1: Opening remarks...4748[00:45] Speaker 2: Response...4950[01:30] Speaker 1: Follow-up...51```5253### Guidelines5455- **Timestamps**: Include approximate timestamps in `[MM:SS]` format at natural breaks (new speakers, topic changes, pauses).56- **Speaker identification**: Label distinct speakers as "Speaker 1", "Speaker 2", etc. If names are mentioned, use them after first identification.57- **Filler words**: Omit excessive filler words (um, uh, like) unless they carry meaning.58- **Inaudible sections**: Mark unclear audio as `[inaudible]` or `[unclear]`.59- **Non-speech sounds**: Note significant sounds like `[laughter]`, `[applause]`, `[music]`.60- **Summary**: Always include a brief summary at the top for quick reference.61- **Frontmatter**: Link back to the source file using a wikilink.6263## Tips6465- For long recordings, let the user know the transcription may be partial due to the 20 MB size limit. Suggest splitting large files with an external tool.66- If the user asks to "transcribe the recording in this note", use `read_file` on the current note first to find embedded audio/video links (e.g., `![[recording.mp3]]`), then `read_file` on the linked file.67- For meeting notes, suggest adding attendees and action items sections after the transcript.68- For podcasts or interviews, suggest adding a "Key Topics" section with timestamps.